logistics.m

凸优化程序包

% Figure 7.1: Logistic regression
% Section 7.1.1
% Boyd & Vandenberghe, "Convex Optimization"
% Original by Lieven Vandenberghe
% Adapted for CVX by Argyris Zymnis - 01/31/06
%
% We consider a binary random variable y with prob(y=1) = p and
% prob(y=0) = 1-p. We assume that that y depends on a vector of
% explanatory variables u in R^n. The logistic model has the form
% p = exp(a'*u+b)/(1+exp(a'*u+b)), where a and b are the model parameters.
% We have m data points (u_1,y_1),...,(u_m,y_m).
% We can reorder the data so that for u_1,..,u_q the outcome is y = 1
% and for u_(q+1),...,u_m the outcome is y = 0. Then it can be shown
% that the ML estimate of a and b can be found by solving
%
% minimize sum_{i=1,..,q}(a'*u_i+b) - sum_i(log(1+exp(a'*u_i+b)))
%
% In this case we have m = 100 and the u_i are just scalars.
% The figure shows the data as well as the function
% f(x) = exp(aml*x+bml)/(1+exp(aml*x+bml)) where aml and bml are the
% ML estimates of a and b.

randn('state',0);
rand('state',0);

% Generate data
a =  1;
b = -5 ;
m= 100;

u = 10*rand(m,1);
y = (rand(m,1) < exp(a*u+b)./(1+exp(a*u+b)));
plot(u,y,'o')
axis([-1,11,-0.1, 1.1]);

% Solve problem
%
% minimize  -(sum_(y_i=1) ui)*a - b + sum log (1+exp(a*ui+b)

U = [ones(m,1) u];
cvx_begin
    variables x(2)
    maximize(y'*U*x-sum(logsumexp([zeros(1,m); x'*U'])))
cvx_end

% Plot results and logistic function

ind1 = find(y==1);
ind2 = find(y==0);

aml = x(2);  bml = x(1);
us = linspace(-1,11,1000)';
ps = exp(aml*us + bml)./(1+exp(aml*us+bml));

dots = plot(us,ps,'-', u(ind1),y(ind1),'o',...
     u(ind2),y(ind2),'o');

axis([-1, 11,-0.1,1.1]);
xlabel('x');
ylabel('y');